Capability is not feasibility.
Our planning rests on one lesson from earlier simulation work: a robot having the right parts does not mean it can do the job. We check feasibility before anything is priced, and we publish the grade of every number we show.
Thirteen rules constrain every design decision. These nine matter most for the marketplace.
| No. | Rule | Why |
|---|---|---|
| P2 | Physics cannot be voted on. Every limit is tagged as a physical fact, a negotiated policy, or a physical quantity with a negotiated margin. | No contract, price or vote can make something physically impossible allowed. A valid signature proves who sent a request, not that it can be done. |
| P3 | Safety is local, and it comes first. The authority to admit, stop or fall back sits on the robot. | A robot must stop safely when the connection drops, a payment fails or a contract is disputed. Finishing the task always ranks last. |
| P4 | Keep the states apart: capability, availability, readiness, priority, feasibility, admission, outcome, verified outcome. | Merging capability with feasibility produced the first real bug in the simulation work. |
| P5 | Task ≠ skill ≠ strategy ≠ implementation. Only the first three travel between robots. | A trajectory that is right for one body can be dangerous for another. |
| P6 | Rule out the infeasible before optimizing. Cheap checks run before expensive ones. | A bid that cannot be admitted is never the cheapest bid, and economic pressure must not leak into safety. |
| P7 | Safety-critical changes are never self-authorized. Learned improvements are proposals with lineage, metrics, scope and negative results. | A system that improves itself without a gate optimizes whatever is measured, including the measurement. |
| P8 | Proposed ≠ proven. Every claim carries a status; simulation is not hardware. | A status rises only after a finished experiment that does not contradict itself. |
| P9 | Reuse before inventing. Build on existing standards and add only what they lack. | A novelty claim that turns out false destroys credibility with partners and certifiers. |
| P13 | Prove the smallest complete loop first: publish, bid, admit, execute, verify, improve. | Whether that loop works across machines with different owners is the unproven core of the venture. |
One unchanged pipeline moved two different arms to a target and refused a third before solving.
The work started from an earlier concept and simulation prototype, which we took over and re-examined. All results are from simulation; nothing has run on hardware.
| Arm | Joints | Reach from base, m | Result | Error at target, mm | Evidence |
|---|---|---|---|---|---|
| A | 2 revolute | 0.95 | Reached | 6.62 | Simulated |
| B | 3 revolute | 0.95 | Reached | 2.94 | Simulated |
| C | 2 revolute | 0.55 | Refused as geometrically infeasible, before the solver ran | — | Simulated |
target (0.40, 0, 0.50) mtolerance 20 mmactuators limited to 10 N·mall joints in one planerun E3
The lesson that became the fit check
The first version passed arm C on a structural check and failed it only later. Two fixes followed: check reach before solving, and make the tool tip an explicit, verified reference frame instead of the last joint.
The same rule now applies to whole jobs: decide which robots can physically do which parts before anyone compares prices.
| Not shown yet | Known limits of the record |
|---|---|
| Categorically different bodies: mobile manipulators, humanoids, 6-axis arms | The 90 N push acted for one simulation step: an impulse of about 0.375 N·s |
| Contact, force and friction | The recovery flag never triggered; the error stayed inside its bound |
| More than one robot, real hardware, real-time timing | The fidelity score is pass or fail, not a measure |
| Anything at the exchange level: tasks, bids, evidence, reuse | Run E3 cannot be re-run yet: its main script is missing from the archive |
Every subsystem is at TRL 3 or below. The first pilots are meant to take the planner and proof of work into a real building.
General-purpose robots will make robot time tradable, once a new task can be loaded quickly.
That is where an open way to describe and transfer skills becomes the moat. We park it until the marketplace works, and will build it on existing standards rather than against them.
Working name
Universal Machine Protocol (UMP): a thin protocol for the exchange itself, covering task, offer, booking, admission and evidence.
"Machine" widens the scope beyond robots to forklifts, cranes and drones. The name is descriptive and will not be a trademark; the brand is protected through Cyberstrata.
| Area | Standards and projects we build on |
|---|---|
| Dispatch across fleets | Open-RMF, VDA 5050, FIPA Contract Net |
| Robot description | URDF, ROS 2 |
| Industrial integration | OPC UA, ISA-95, Asset Administration Shell (IEC 63278) |
| Safety | ISO 10218, ISO/TS 15066, ISO 13849 |
| Skill and capability contracts | RFL, SkillCrate, ECM |
Every number on this site carries its grade. No grade, no number.
| Grade | Meaning | Example on this site |
|---|---|---|
| Measured | Measured by us on real hardware or with real customers | None yet |
| Simulated | Output of a simulation or model run | 2.94 mm error at target, arm B |
| Estimate | Calculated by us from stated assumptions | €1800–4500 per building per month |
| Assumption | An input we have not verified yet | 15 % commission |
| Reported | Published by a third party; the source is linked | 25–40 % robot share of a cell's cost |
| Target | A goal with a date or a deadline | 3 paid pilot jobs by week 10 |